gemini omni flash dropped recently. we ran it against the current top 4 text-to-video models on @arena – all one-shot, same prompt, via @aimlapi
the prompt: a single continuous fpv drone shot. opens low inside a moving subway car, weaves between passengers, banks around a pole, threads into the next carriage, dives out an open window onto an elevated track, then climbs alongside a skyscraper into a golden-hour skyline. cinematic, no cuts, no text
results:
omni flash
• generation time: 31 sec
• length: 10 sec
• cost: $1.32
happy horse 1.1
• generation time: 2 min 57 sec
• length: 10 sec
• cost: $2.32
seedance 2.0
• generation time: 1 min 37 sec
• length: 5 sec
• cost: $2.00
veo 3.1
• generation time: 2 min 07 sec
• length: 5 sec
• cost: $4.16
the interesting part:
- only seedance 2.0 and happy horse 1.1 (chinese labs!) read it as a pov shot and kept the drone out of frame. google's models both spawned a visible drone flying through the scene
- seedance was the only output with no lags or hallucinations, but it ran just 5s
- omni flash gave the weakest result for the lowest price. its drone flies straight through objects
- veo 3.1 felt the most natural – fish-eye effect and all
the takeaway: you can see why it's called flash. omni's videos generate fast and cheap, but the quality is poor
which model won?
We’re shipping 2 major releases:⁰
— Google DeepMind (@GoogleDeepMind) June 30, 2026
🔘 Nano Banana 2 Lite: our fastest and cheapest Gemini Image model
🔘 Gemini Omni Flash: now available via the Gemini API and in @GoogleAIStudio to help developers generate and edit high-quality videos. pic.twitter.com/fqB2sA5Xyl
Nick Trenkler